Xgesvdr

cuSOLVER Singular Value Decomposition example

Description

This code demonstrates a usage of cuSOLVER Xgesvdp 64-bit functions for using cusolverDnXgesvdr. The code shows how to compute the rank-2 approximation of the 5x5 dense matrix A:

A = | 0.7640 | 0.0369 | 0.4201 | 0.5033 | 0.5832 |
    | 0.6141 | 0.8596 | 0.3920 | 0.9297 | 0.1158 |
    | 0.8172 | 0.6758 | 0.1265 | 0.2121 | 0.3983 |
    | 0.4204 | 0.4559 | 0.9025 | 0.6392 | 0.2149 |
    | 0.0344 | 0.0207 | 0.2307 | 0.5812 | 0.0054 |

The following code uses two steps:

Step 1: compute the approximated rank-2 decomposition of A

Step 2: check accuracy of singular values

Supported SM Architectures

All GPUs supported by CUDA Toolkit (https://developer.nvidia.com/cuda-gpus)

Supported OSes

Linux
Windows

Supported CPU Architecture

x86_64
ppc64le
arm64-sbsa

CUDA APIs involved

• cusolverDnXgesvdr_bufferSize API
• cusolverDnXgesvdr API

Building (make)

Prerequisites

• A Linux/Windows system with recent NVIDIA drivers.
• CMake version 3.18 minimum
• Minimum CUDA 11.1 toolkit is required.

Build command on Linux

$ mkdir build
$ cd build
$ cmake ..
$ make

Make sure that CMake finds expected CUDA Toolkit. If that is not the case you can add argument -DCMAKE_CUDA_COMPILER=/path/to/cuda/bin/nvcc to cmake command.

Build command on Windows

$ mkdir build
$ cd build
$ cmake -DCMAKE_GENERATOR_PLATFORM=x64 ..
$ Open cusolver_examples.sln project in Visual Studio and build

Usage

$  ./cusolver_Xgesvdr_example

Sample example output:

2, 2
A = (matlab base-1)
0.76 0.04 0.42 0.50 0.58
0.61 0.86 0.39 0.93 0.12
0.82 0.68 0.13 0.21 0.40
0.42 0.46 0.90 0.64 0.21
0.03 0.02 0.23 0.58 0.01
=====
m = 5, n = 5, rank = 2, p = 2, iters = 2
after Xgesvdr: info = 0
S_ref[0]=2.365392  S_gpu=[0]=2.365392  AbsErr=4.975184E-09  RelErr=2.103323E-09
S_ref[1]=0.811178  S_gpu=[1]=0.811178  AbsErr=2.060349E-09  RelErr=2.539947E-09

max_err = 0.000000E+00, max_relerr = 0.000000E+00, eps = 1.000000E-08
Success: max_relerr is smaller than eps